UKB as LD reference panel to do fine-mapping in loci from EUR ancestry GWAS meta-analysis

[AMCalejandro]

Hi,

Thanks for this useful tool.

I am performing fine-mapping on a GWAS meta-analysis of PD European ancestry samples only. I a concerned about the false positives from subtle differences between the LD ref panel and the target sample population.

Considering the sample size of UKB as well as keeping in mind that when performing some ancestry clustering, British ancestry ppl consistently overlap with the rest of European ancestry samples, I wonder if this setting is robust enough to be confident on the selection of the reference panel.

The data to derive the GWAS meta-analysis was linked to stringent ancestry checks

[jdblischak]

I a concerned about the false positives from subtle differences between the LD ref panel and the target sample population.

One option is to use a tool like DENTIST from Chen et al. 2021. It removes SNPs with GWAS test statistics that don't closely match their predicted values based on the test statistics of neighboring SNPs and the reference LD measurements.

[omerwe]

@AMCalejandro that's a good question, and we spent a lot time trying to address it when writing the PolyFun manuscript. There are no easy answers unfortunately... I encourage you to look at Table 3 in the PolyFun paper (and the text referencing it) as well as the Supplementary Note, which address these questions in detail.

@jdblischak that's an interesting idea, but removing SNPs from fine-mapping is unfortunately problematic, as it runs into risk of removing the actual causal SNP...

[jdblischak]

that's an interesting idea, but removing SNPs from fine-mapping is unfortunately problematic, as it runs into risk of removing the actual causal SNP...

Of course I agree with you that in general it is not ideal to remove SNPs prior to fine-mapping. But in the very non-ideal situation of using an out-of-sample LD reference to fine-map the results of a GWAS meta-analysis, I don't think it is advisable to include SNPs that are clearly mismatched between the study population and LD reference panel. DENTIST uses an iterative procedure to remove as few SNPs as possible.

Chen et al. 2021 specifically performed a simulation to demonstrate improved fine-mapping performance post-filtering with DENTIST. I'd be keen to hear any potential shortcomings of these results. Skimming the peer review file, the DENTIST authors added the iterative procedure specifically to address reviewer comments and reduce the risk of removing causal variants.

[omerwe]

@jdblischak I agree! Thanks for informing me about this tool.